Quickstart

This guide walks a business problem all the way to a live, callable prediction — entirely inside the Studio UI. No SDK, no code, until the very last step where your application calls the finished endpoint.

You'll need an account and a workspace. If you don't have one yet, your organization administrator can invite you, or you can register a new organization from the sign-in screen. Once signed in, pick your organization and workspace in the switcher at the top of the screen — everything you create lives inside that workspace.

1. Create a Solution

Go to Solutions and click Create solution. Name it after the business problem — Asset Intelligence, Customer Churn — and describe the outcome you want. This is the container for everything that follows. → Solutions

2. Add a Capability

Open the solution and add a capability. Choose its type (classification, regression, forecasting, …), declare the input contract — the fields it will receive — and set the acceptance criteria: the quality bar a model must clear before it can ever go live. → Capabilities · Acceptance criteria

3. Bring data

Go to Datasets and upload a CSV — or set up a connector under Connectors to pull data from Postgres, MySQL, or S3 on a schedule. The platform validates the data and records it as a version. → Datasets

4. Define a Feature Set

From your dataset version, define the model-ready features (and the target column, if your capability predicts a known answer). The same definition is used at training time and at prediction time, so the model always sees data the way it was trained. → Feature sets

5. Train

On the capability page, start a training run: point it at your dataset and feature set versions and pick an algorithm — a good default is provided. Watch the run under Training; it is recorded as an experiment you can revisit and compare. → Training · Experiments

6. Evaluate and promote

When training succeeds, the candidate model is evaluated on held-out data and checked against your acceptance criteria. A model that passes can be promoted; a model that fails is blocked with the failing rule spelled out — no exceptions, no manual overrides. → Models & evaluation

7. Deploy

Go to Deployments and create a deployment for the capability. The platform starts it and exposes a stable endpoint that stays the same across every future model version. → Deployments

8. Try it

Open the deployment's detail page and use the built-in playground to send a test prediction straight from the browser — you'll see the shaped output and the round-trip latency without writing any code.

9. Connect your application

When you're ready to integrate, your application calls the same endpoint:

curl -X POST https://your-host/api/v1/predict/CAPABILITY_KEY \
  -H 'Authorization: Bearer YOUR_TOKEN' \
  -H 'Content-Type: application/json' \
  -d '{"inputs":{"age":30,"income":50000}}'

Create the token under your profile menu → API Tokens (how), and see SDKs & API clients for Python / TypeScript / Go helpers.

10. Operate

Once live, the deployment appears under Monitoring. The platform tracks health and latency, watches for data drift, and — if you set up a retraining policy — proposes a fresh model that must pass the same gate as the first one.

That's the whole loop: problem → data → model → endpoint → operate. Every step is versioned and audited, so you can always see how you got here and roll back if you need to.